The Soil Conservation Service (SCS) curve number (CN) estimates of direct runoff from rainfall for semiarid catchments can be inaccurate. Investigation of the Walnut Gulch Experimental Watershed (WGEW) (Southeastern Arizona) and its ten nested catchments determined that the inaccuracy is due to the original SCS ratio (λ) of 0.2 between initial abstraction and maximum potential retention. Sensitivity analyses indicate that runoff estimation can be very sensitive to the initial abstraction ratio, especially for relatively low rainfall amount and for watersheds covered by deep, coarse, and porous soil, conditions that dominate many semiarid watersheds worldwide. Changing the ratio of initial abstraction to the maximum potential retention to optimal values ranging from 0.01 to 0.53 for different Walnut Gulch catchments improved runoff estimates. The greater the channel area and the finer the soil, the smaller the initial abstraction ratio is. The variation of the initial abstraction ratio for the WGEW is due to the variation of maximum potential retention and initial abstraction, which are channel area and soil-dependent parameters. The greater the channel area, the higher the maximum potential retention S is, and the coarser the soil, the larger the initial abstraction Ia is. In addition, the effect of initial abstraction ratio on runoff estimation increases with decreasing CN. Thus, impacts of initial abstraction ratio on runoff estimation should be considered, especially for semiarid watersheds where the CN is usually low. Copyright © 2012 John Wiley & Sons, Ltd.